Authors:
(1) Simone Silvestri, Massachusetts Institute of Technology, Cambridge, MA, USA;
(2) Gregory Wagner, Massachusetts Institute of Technology, Cambridge, MA, USA;
(3) Christopher Hill, Massachusetts Institute of Technology, Cambridge, MA, USA;
(4) Matin Raayai Ardakani, Northeastern University, Boston, MA, USA;
(5) Johannes Blaschke, Lawrence Berkeley National Laboratory, Berkeley, CA, USA;
(6) Valentin Churavy, Massachusetts Institute of Technology, Cambridge, MA, USA;
(7) Jean-Michel Campin, Massachusetts Institute of Technology, Cambridge, MA, USA;
(8) Navid Constantinou, Australian National University, Canberra, ACT, Australia;
(9) Alan Edelman, Massachusetts Institute of Technology, Cambridge, MA, USA;
(10) John Marshall, Massachusetts Institute of Technology, Cambridge, MA, USA;
(11) Ali Ramadhan, Massachusetts Institute of Technology, Cambridge, MA, USA;
(12) Andre Souza, Massachusetts Institute of Technology, Cambridge, MA, USA;
(13) Raffaele Ferrari, Massachusetts Institute of Technology, Cambridge, MA, USA.
Table of Links
5.1 Starting from scratch with Julia
5.2 New numerical methods for finite volume fluid dynamics on the sphere
5.3 Optimization of ocean free surface dynamics for unprecedented GPU scalability
6 How performance was measured
7 Performance Results and 7.1 Scaling Results
9 Acknowledgments and References
8 Implications
By developing a new model from scratch specifically for GPUs, and wielding a handful of key ocean-model-specific innovations, Oceananigans achieves 9.9 SYPD at 10 km resolution using less than 1% of the resources of current state of the art supercomputers. This achievement means that most climate model runs submitted to IPCC will be able use 10 km ocean models — precipitating a step change in the accuracy of climate prediction.
At scales between 10–100 km, macroscale ocean turbulence exerts a key control on ocean carbon and heat uptake. However, attempts to accurately parameterize this key process in coarse resolution models have frustrated generations of oceanographers. The inadequacies of macroscale parameterizations are associated with major biases and uncertainty in climate predictions [33, 39]. At resolutions of 10 km, the need for macroscale turbulence parameterization is eliminated, and ocean simulations capture key ocean features such as sharp sea surface temperature gradients supporting the formation of marine stratus clouds above narrow eastern boundary currents like the California and Benguela Current [28], and changes in the meridional overturning circulation due to the effect of Antarctic meltwater on deep convection in austral winter [26].
Additionally, by achieving 0.95 SYPD at 1.7 km resolution, we pave the way for decadal ocean simulations of the ocean “submesoscale” — the ocean analogue to atmospheric weather — which exhibits hourly fluctuations, high spatial and seasonal variability, and which exerts a strong control on ocean air-sea fluxes, biological productivity and fish stocks [46]. The granularity and accuracy provided by 1.7 km resolution is further required to plan local mitigation strategies and predict local extreme events.
Third, the unparalleled speed of execution and memory efficiency of Oceananigans allows global computations at never-before-seen sub-kilometer resolutions. The capacity for ultra-high-resolution simulations aligns with current advancements in resolution of ocean sampling platforms from satellites [32, 12] to fleets of floats and drones. While this wealth of data is likely to provide new insights and scientific knowledge about the nature of small scale processes, global high-resolution ocean simulations will be needed to explore their impact on global climate scales.
Finally, our results pave the way for marked increase in energy efficiency of climate simulations. The very reason to develop climate models, as stated by the Coupled Model Intercomparison Project (CMIP), for example, is to provide the necessary information to effectively reduce emissions and mitigate the effects of global warming — while, counterproductively, the carbon footprint of climate simulations that contribute to CMIP increases rapidly. Oceananigans’ achievements represent a milestone towards decreased energy consumption by climate modeling efforts.
This paper is available on arxiv under CC BY 4.0 DEED license.
